Yes, you can use Nemotron 3.5 Super VL OpenCode setups for free right now: open OpenCode, click the model drop-down, and pick "Nemotron 3.5 Super VL". That is it. No API key juggling, no download. This page shows you how I use it, the build I tested it on, a local route for big machines, and where it falls short.
📺 Watch: NVIDIA Nemotron 3.5 Super VL Is INSANE! (FREE AI Model)
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
I tested this inside OpenCode in my own Agentic Operating System. It worked first time. It is not a frontier model and it is not going to replace Opus 5.5, but for a free, open, fast coding model that can also see images, it is a very useful thing to have one click away.
If you want my full Agent OS with OpenCode, Hermes, OpenClaw and Claude already set up side by side, it comes inside the AI Profit Boardroom ($69/mo).
How to pick Nemotron 3.5 Super VL in OpenCode
Here are the steps, start to finish:
- Open OpenCode in your project folder.
- Open the model drop-down (the model picker).
- Find "Nemotron 3.5 Super VL" in the list and select it.
- Type your first prompt and hit enter.
That is the whole setup. NVIDIA lists OpenCode as one of the launch harnesses for this model, which is why it shows up there ready to go. For the bigger picture on what is free and where, see my guide to using Nemotron 3.5 Super VL free.
Quick model background if you are new to it: Nemotron 3.5 Super VL is NVIDIA's open multimodal model for agents. 120B total parameters, mixture-of-experts, 12B active per token. It reads text, images, documents and video frames and writes text back. It handles reasoning and tool use in one model. The full breakdown is in my Nemotron 3.5 Super VL overview.
My to-do app test
I wanted a simple, honest test. So I gave it this prompt:
Build a beautiful to-do list app. Make it entertaining, engaging.
That is a vague prompt on purpose. Most people prompt like that, so I want to see what a model does with it.
The build came out working. A real to-do app I could click through, add tasks to and tick off. For a free model running inside OpenCode with zero setup, that is a solid result.
Was it the best build I have ever seen? No. Frontier models still give you more polish and fewer rough edges on the first try. But the point is this: you can get a working app built on a free model in a couple of minutes. That changes what is worth trying.
Want to try the same thing? Copy my prompt, swap "to-do list app" for whatever you need, and see what you get. Then iterate. Ask for changes in plain English, one at a time.
📺 Watch: OpenClaw + Nvidia Nemotron 3 Super + Ollama is INSANE!
Running it locally with Ollama and OpenCode
If you have a big-memory machine, there is a second route. In Ollama, the model is listed as nemotron-3.5-super, and it is the same vision model. Ollama can launch OpenCode wired straight to it:
ollama launch opencode --model nemotron-3.5-super
Be honest with yourself about the hardware. The Ollama download is roughly 80-90GB, so you need that much memory free, like a 128GB DGX Spark or a big workstation. On a normal laptop, stick with the free hosted model in the OpenCode drop-down. It is the faster and simpler option for most people.
📺 Watch: New FREE Nvidia Nemotron 3 Super + Ollama
Best coding and vision jobs for Nemotron 3.5 Super VL OpenCode work
The edge here is vision. Plenty of free coding models can write code. Fewer can look at a picture and code from it. So give it the jobs that start with an image or a file.
Screenshot to UI
Take a screenshot of a layout you like, or sketch one, and ask OpenCode to rebuild it. Something like: "Here is a screenshot of a pricing section. Build this as a simple HTML page that matches the layout." It reads the image and writes the code.
Reading error screenshots
This is the one I like most for beginners. You get a weird error in the browser or a terminal and do not know what it means. Screenshot it, drop it in, and ask: "What is this error and how do I fix it in this project?" No more copying long messages by hand.
Reading docs and PDFs
Got an API reference as a PDF, or a spec doc from a client? Give it the document and ask it to build to that spec. It handles documents natively. NVIDIA also reports strong OCR results, placing it among the top open models on OCRBench v2.
Charts and dashboards
Per NVIDIA, it scores 80.8% on CharXiv chart reasoning. That makes it handy when you want to turn a chart screenshot into a working chart in code, or check that your dashboard shows the right numbers.
On the coding side, NVIDIA says it beats Nemotron 3 Super on SWE-bench and Terminal-Bench setups, though they do not give numbers for those. All of these benchmarks are NVIDIA's own and nobody has independently replicated them yet, so test on your own tasks. You can see how I score models on real work in Goldie Bench.
If you want help working out which jobs in your business should run on free models and which need a frontier one, book a free AI strategy session.
The limits, straight up
Here is the honest picture so you do not get burned.
- It is not frontier. For your hardest builds, big refactors or anything where a bug costs real money, keep a frontier model on the job.
- Free means limits. Free models in OpenCode can run out or get busy. When that happens, do not stop. Switch to another free model in the drop-down. I named Big Pickle as one option in my video. Or move the task over to Hermes.
- Vague prompts give vague builds. My to-do test worked, but the more specific your prompt, the better the result. Tell it the features, the style and the stack.
- Check its work. Run the app. Click everything. Ask it to fix what breaks. That loop is where the real value comes from.
How it fits in my stack
I do not use one model for everything. My Agent OS runs OpenCode, Hermes, OpenClaw and Claude side by side, and I swap models in and out depending on the job. New open models like this one slot straight in.
Nemotron 3.5 Super VL in OpenCode is my go-to for quick, free, picture-first coding jobs. Frontier models handle the heavy lifting. If you want to see how all the pieces connect, read my Agent OS guide.
And if you are comparing OpenCode against other coding workspaces, my OpenWork AI write-up covers another way to run agents on your desktop.
The fastest way to get my exact setup is inside the AI Profit Boardroom. You get the Agent OS zip, a video tutorial, a 30-day roadmap, updates as new models drop, four weekly coaching calls and daily tutorials. Over 3,400 members are already using it.
FAQ
Is Nemotron 3.5 Super VL free in OpenCode?
Yes. Pick "Nemotron 3.5 Super VL" in the OpenCode model drop-down and start prompting. Free access can have limits, so switch to another free model if you run out.
Can Nemotron 3.5 Super VL read screenshots in OpenCode?
Yes. It is a vision-language model, so it can read screenshots, error messages, PDFs and charts. That is the main reason to pick it over other free coding models.
What is the Ollama command to run it with OpenCode?
ollama launch opencode --model nemotron-3.5-super. You need roughly 80-90GB of memory for the local model, so most people should use the hosted version in the drop-down instead.
Should I use it instead of a frontier model for coding?
Not for your hardest work. Use it for fast, free, image-first jobs and keep a frontier model for big builds. If you want a plan for your own setup, book a free AI strategy session.
That is the full playbook for Nemotron 3.5 Super VL OpenCode work: pick it in the drop-down, give it jobs that start with a picture, and switch models when you hit a limit.
